MULTi-modal Integration in Pancreas cancer using machine Learning (MULTIPL): Personalized predictions of outcomes across the spectrum of disease and treatments

Objective: Pancreas cancer is highly aggressive, and the only potential cure is surgical resection. Among the 11% of patients eligible for surgery, most see the disease recur within short years. Patients with advanced disease typically survive less than a year after diagnosis. Chemotherapy options are limited, and patients frequently see their disease grow on the first CT scan after starting therapy. Fewer than half of patients with advanced disease are able to receive second-line therapy. Given the aggressive nature of this cancer, making the best treatment decision at each opportunity is critical. Recent discoveries in genomic analysis of pancreas cancer improved our understanding of its underlying biology, but these advances have not yet impacted decision-making in the clinic. We propose to use artificial intelligence to combine different types of data to help select the best treatment for each patient.

Methods: Our team has extensive experience in using genomic technology to discover the biology of pancreas cancer. In the COMPASS trial, patients with advanced pancreas cancer had a biopsy of their cancer taken and analyzed to reveal its whole genome (all mutations in the cancer) and transcriptome (level of expression of each of these genes). We know that the genomic and transcriptomic data provide complementary information, and other data such as the pathology (microscope) images, medical imaging data and clinical information are also important in understanding pancreatic cancer. In the current study, we plan on using state-of-the-art artificial intelligence, that is to build a computer model that integrates information from all the rich and diverse data sources we have developed to predict which cancers will grow or respond to different therapies. We will then test our predictive computer model in new ongoing studies called PASS-01, NeoPancOne, Prosper-PANC and ABLATE, to validate the accuracy of our system.

Impact and relevance to cancer: In our study, we will develop a computer model capable of predicting the most aggressive forms of pancreas cancer from the mutations detected in genomic analyses. Our hope is that this algorithm helps to determine which patients benefit most from surgery and which ones would instead be hurt by interrupting chemotherapy too soon to undergo surgery only to see their cancer come back rapidly. Our data and algorithms will assist scientists across Canada, and in the world, in targeting treatment in pancreas cancer to the precise needs of individual patients.

Quotes

“I am grateful to receive the MOHCCN HI&DS award to support my project, which leverages machine learning to integrate various data modalities for a deeper understanding of pancreatic cancer biology and prognosis. Pancreatic cancer is an aggressive disease, and this award provides a fantastic opportunity to work towards improving patient outcomes and developing precision tools for tailored treatments.”

• Dr. David Henault, HI&DS Awardee

“There are promising developments in oncology demonstrating the power of machine learning in improving care for various types of cancer, yet pancreatic cancer remains behind. We are uniquely positioned to address this gap due to our rich, deeply characterized pancreatic cancer cohorts. I believe this project will yield valuable insights from our extensive data and represent a significant advance towards precision medicine for our pancreatic cancer patients.”

• Dr. Robert Grant, mentor